Re: [PATCH v4] KVM: VMX: Fix enable VPID even if INVVPID is not exposed in vmx capability

[David Hildenbrand]

On 23.03.2017 03:39, Wanpeng Li wrote:
> From: Wanpeng Li <wanpeng.li@xxxxxxxxxxx>
> 
> This can be reproduced by running L2 on L1, and disable VPID on L0 
> if w/o commit "KVM: nVMX: Fix nested VPID vmx exec control", the L2 
> crash as below:
> 
> KVM: entry failed, hardware error 0x7
> EAX=00000000 EBX=00000000 ECX=00000000 EDX=000306c3
> ESI=00000000 EDI=00000000 EBP=00000000 ESP=00000000
> EIP=0000fff0 EFL=00000002 [-------] CPL=0 II=0 A20=1 SMM=0 HLT=0
> ES =0000 00000000 0000ffff 00009300
> CS =f000 ffff0000 0000ffff 00009b00
> SS =0000 00000000 0000ffff 00009300
> DS =0000 00000000 0000ffff 00009300
> FS =0000 00000000 0000ffff 00009300
> GS =0000 00000000 0000ffff 00009300
> LDT=0000 00000000 0000ffff 00008200
> TR =0000 00000000 0000ffff 00008b00
> GDT=     00000000 0000ffff
> IDT=     00000000 0000ffff
> CR0=60000010 CR2=00000000 CR3=00000000 CR4=00000000
> DR0=0000000000000000 DR1=0000000000000000 DR2=0000000000000000 DR3=0000000000000000 
> DR6=00000000ffff0ff0 DR7=0000000000000400
> EFER=0000000000000000
> 
> Reference SDM 30.3 INVVPID:
>  
> Protected Mode Exceptions
> #UD 
>   - If not in VMX operation.
>   - If the logical processor does not support VPIDs (IA32_VMX_PROCBASED_CTLS2[37]=0).
>   - If the logical processor supports VPIDs (IA32_VMX_PROCBASED_CTLS2[37]=1) but does 
>     not support the INVVPID instruction (IA32_VMX_EPT_VPID_CAP[32]=0).
> 
> So we should check both VPID enable bit in vmx exec control and INVVPID support bit 
> in vmx capability MSRs to enable VPID. This patch adds the guarantee to not enable VPID
> if INVVPID is not exposed in vmx capability MSRs.

You should change the subject and this description to also take the
single/global capability into account.

> 
> Reviewed-by: David Hildenbrand <david@xxxxxxxxxx>
> Cc: Jim Mattson <jmattson@xxxxxxxxxx>
> Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx>
> Cc: Radim KrÄmÃÅ <rkrcmar@xxxxxxxxxx>
> Signed-off-by: Wanpeng Li <wanpeng.li@xxxxxxxxxxx>
> ---
>  arch/x86/kvm/vmx.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> index 8795a70..8925c76 100644
> --- a/arch/x86/kvm/vmx.c
> +++ b/arch/x86/kvm/vmx.c
> @@ -1239,6 +1239,11 @@ static inline bool cpu_has_vmx_invvpid_global(void)
>  	return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
>  }
>  
> +static inline bool cpu_has_vmx_invvpid(void)
> +{
> +	return vmx_capability.vpid & VMX_VPID_INVVPID_BIT;
> +}
> +
>  static inline bool cpu_has_vmx_ept(void)
>  {
>  	return vmcs_config.cpu_based_2nd_exec_ctrl &
> @@ -6518,8 +6523,10 @@ static __init int hardware_setup(void)
>  	if (boot_cpu_has(X86_FEATURE_NX))
>  		kvm_enable_efer_bits(EFER_NX);
>  
> -	if (!cpu_has_vmx_vpid())
> +	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
> +		!cpu_has_vmx_invvpid_single() || !cpu_has_vmx_invvpid_global())

According to Jim

"KVM's vpid_sync_context() assumes that at least one of
{VMX_VPID_EXTENT_SINGLE_CONTEXT, VMX_VPID_EXTENT_ALL_CONTEXT} ..."

So we should only require one of these two, not both.

if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))

>  		enable_vpid = 0;
> +
>  	if (!cpu_has_vmx_shadow_vmcs())
>  		enable_shadow_vmcs = 0;
>  	if (enable_shadow_vmcs)
> 


-- 

Thanks,

David